Zyphra's ZUNA 1.1 Rebuilds Messy Brain Signals With 90% Electrode Dropout

ZUNA 1.1 is the latest version of Zyphra's open EEG foundation model, and it is a meaningful step up from its predecessor. The update brings four concrete upgrades that make the model far more practical for real-world EEG data: variable-length input windows, four new realistic dropout training schemes, quality-aware preprocessing, and a browser-based playground on Zyphra Cloud that requires no GPU or coding knowledge. The weights are Apache 2.0 licensed and free to use.

The problem no one talks about

While scalp EEG data is widely available and noninvasive, it is often messy and incomplete. EEG datasets frequently include partially corrupted sessions, and a few bad channels can render an otherwise valuable recording unusable. The traditional fix has been spherical-spline interpolation, a geometric method built into MNE that smooths signals across the scalp. Spherical splines perform reasonably well when only a few electrodes fail, but as dropout increases, whether from motion artifacts, hardware faults, or low-density headsets, signal quality deteriorates.

Unlike text, images, or audio, the EEG domain still lacks general and powerful foundation models. Data fragmentation is a major reason: EEG datasets are typically small, collected under different protocols, and distributed across many institutions. Different datasets use varying numbers of channels and inconsistent electrode positions, and most deep learning models are trained on fixed channel montages, making them fail when applied to new datasets or recording conditions.

What ZUNA actually is

ZUNA is a 380M-parameter masked diffusion autoencoder trained to perform masked channel infilling and superresolution for arbitrary electrode numbers and positions in EEG signals. A diffusion autoencoder is a model that compresses data into a compact internal representation (the encoder) and then reconstructs it by iteratively denoising random noise into a clean signal (the decoder). ZUNA leverages a diffusion autoencoder architecture based on a transformer backbone. An encoder maps EEG signals to a shared latent space, and a decoder reconstructs EEG signals from latents. It was trained with a masked reconstruction loss and a heavy dropout scheme, enabling ZUNA to denoise existing channels and predict new ones during inference.